1. Field of the Invention
The present invention concerns a method and a device for communicating information.
It applies in particular to an asynchronous packet switched network, making it possible notably to interconnect a small number of items of multimedia equipment, whilst providing them with different service qualities for exchanging data.
2. Description of the Related Art
The behavior of data traffic generated by an item of multimedia equipment (or application) can fall within two main service classes, themselves each divided into sub-classes. Thus the traffic can be either elastic (it adapts easily to changes in transmission conditions), or real time. If it is elastic, it can either be interactive in nature, and therefore sensitive to the transfer time, or can correspond to a massive transfer of a large volume of data and therefore sensitive to the passband. If it is a case of real-time traffic, either it is acceptable to lose certain information packets in order to give priority to the transmission time, and the traffic is then referred to as “predictive”, or it is preferable to have a moderate transmission rate but without loss of data, and the traffic is then referred to as “guaranteed”.
The elastic traffic corresponds to traffic of the datagram type. This traffic is said to be elastic since it is capable of adapting to the transmission conditions without for all that losing its utility. A file transfer can be performed either via a path supporting a transmission rate of 64 kilobits per second or via a path supporting a transmission rate of 2 megabits per second. With elastic traffic, a first basic class concerns the traffic generated by interactive or transactional applications (such as an application of the client-server type), and a second class identifies the data conveyed by block (such as the transfer of files).
The equipment generating a real-time traffic requires a predictive or guaranteed service. For predictive traffic, which gives priority to complying with a high constraint with regard to the transmission time, the application specifies the passband which it is assumed to require, as well as the maximum transmission time which it can accept. The network gives priority to this type of traffic but gets rid of the data for which the transmission time cannot be complied with. It is a case for example of video transmission (in the event of transmission problems, a static image appears) or audio systems of average quality.
The so-called guaranteed service is differentiated from the predictive service by the fact that the network does not intentionally get rid of the data coming from a guaranteed traffic but this may have a variable transmission time according to the predictive traffic with a higher priority or the concurrent guaranteed traffic. It is a case, for example, of high-definition video traffic (using compression techniques). In this case, the network reserves the passband for the application which requires a guaranteed service, but, in order to manage problems of jitter on the data on reception, the application must have temporary data storage capacities (around one second of traffic) in order to re-establish the original behavior of the traffic.
The service quality guarantee required by concurrent applications of different natures is achieved by resolving problems such as the management of resources and the control of traffic. The methods to be used must enable the network to function optimally whilst affording a service quality acceptable to the different applications. This problem has been the subject of many studies for circuit switched networks and for packet switched networks.
In circuit switched networks, the known solution consists of allocating to each connection a passband (channel) which is constant throughout its life. Prior to this, a call acceptance procedure enables the network to know whether it can support the call. Where no channel is available, the call is rejected.
In packet switched networks, where the traffic is, by definition, unpredictable, the variable nature of the transmission rates offers the opportunity of a statistical sharing of the network resources. This optimization of resources unfortunately increases the risks of congestion. These networks can be seen as a succession of queues of limited capacity, whose filling it is necessary to control in order to prevent their saturation, synonymous with loss of packets.
It is known that the traffic can be controlled by a so-called “windowing” mechanism: when a saturation state is detected, the receiver explicitly requests the source to reduce its flow. This is then referred to as flow regulation and control.
Neither of these approaches is sufficient, the first because it leads to an inevitable waste of resources in the network by allocating, to the source, a passband corresponding to its maximum transmission rate, and the second because the ratio of propagation time to transmission time increases considerably when the links are at a very high transmission rate. In systems controlling flow by feedback, between the moment of sending of the notification of congestion and the moment of its reception by the source node, the traffic already in transit in the network can be considered to be definitively lost because of the congestion, unless there are large-capacity memories available for storing all the packets in transit during the congestion state.
Consequently, for rapid switching of packets, purely reactive control methods are insufficient in a high throughput environment. Switching requires elaborate mechanisms in the switching equipment (intermediate nodes in the network) including preventive methods and reactive methods. ATM (Asynchronous Transfer Mode) technology proposes techniques for managing the service quality and controlling congestion based on elaborate mechanisms implemented within the switching equipment. Examples of such mechanisms are described in U.S. Pat. Nos. 5,291,481 and 5,313,454, which also illustrate the complexity and potential cost of implementing such methods.
These methods are therefore not suited to networks comprising a smaller number of items of equipment to be interconnected, where the cost of the communication means must not be too high compared with the cost of the equipment to be interconnected.
Attempts to use on-board ATM switches have been made, such as the one described in the thesis “Real-time distributed architecture based on ATM” by Jean-François Guilaud (INPG). This thesis mentions the use of conventional ATM signaling for managing connections, which represents a complex implementation. However, the complete use of connection management mechanisms must make provision for preventing congestion, which, in order to simplify the use of the switch, are based on a local management of the load information and on the use of queues at the output for each switch.
Strict flow control, at the source, as described in the thesis, provides for the use of a connected mode solely, and checks that the transfer of data has not been exceeded compared with what was predicted when the corresponding connection was established. This can therefore give rise to a poor use of network resources and a lack of flexibility in the system.
Moreover, no simple means is envisaged for reacting to the problem of congestion except one based on the solutions described above, peculiar to the use of ATM, but which remain complex solutions.
In addition, the use of a fixed packet size, in accordance with ATM, gives rise to a fixed loss of the useful throughput per available network link. On the other hand, a packet size which is variable according to the load on the network makes it possible to optimize the useful throughput. The techniques of controlling the size of the packets have already been tested on network architectures of the bus or ring type.
Asynchronous packet switching, as described in the standard IEEE-P1355, is based on a switching technology (cut-through crossbar allowing several simultaneous paths) with a low implementation cost with regard to the switch. This is because the switch in question uses only a minimum of resources for effecting the switching of a packet from an input port to an output port. The transfer of a packet through the switch takes place as soon as the switch has knowledge of the switching information for the said packet (packet header) without awaiting complete reception of all the packet data. There is therefore no management of queue or priority in the intermediate nodes of the path.
Moreover, in order to regulate the problems of contention in access to an input port (or respectively to an output port) of the switch, when several packets are a priori intended to pass via this input port (or respectively output port), a flow control mechanism at the level of the link is implemented, allowing a source to transmit data on the transmission line only when it has obtained authorisation from the destination, that is to say when a group of data previously sent by the source have been acknowledged by the destination.
However, asynchronous packet switching, as known in the state of the art, though it affords attractive implementation costs, does not guarantee different service qualities for concurrent traffic within the same network.
The technologies of the serial bus type propose an alternative to the use of packet switching for interconnecting the peripherals at lower cost. This is because the mechanisms used for sharing resources are simplified because of the unicity of the main resources, in this case the communication medium. This simplicity also entails the drawback of a limited passband: the mean passband available per terminal decreases as a function of the number of terminals connected.
Certain serial bus technologies, such as those in accordance with IEEE standards, reference P1394, defined for the interconnection of multimedia equipment, support the transfer of data in accordance with mainly two service classes using a logic architecture of the bus type. Mechanisms are therefore necessary for arbitrating access to the bus and organising the data transfer. This serial bus technology provides for a resource reservation mechanism. It affords, on the one hand, the so-called isochronic transmission of data, comprising a reservation phase, and, on the other hand, the so-called asynchronous transmission of data, without reservation phase.
The document U.S. Pat. No. 4,914,650 describes a method for organising the transmission of data coming from different queues and for inserting therein an ultra-priority traffic (of the signalling type), coming from a third queue, as well as mechanisms for reacting to the congestion problem in the queues associated with each type of data. It does not propose any solution for limiting the losses of information due to congestion problems and therefore for guaranteeing different service qualities.
The document U.S. Pat. No. 5,621,898 illustrates a mechanism for organising the transmission of data on a bus in accordance with the IEEE-P1394 specifications. It thus describes the sequencing of the packet transmissions, associated with different queues, and the mechanisms controlling access to the bus enabling the available passband to be taken into account for the transmission of additional packets.